Government consultation on AI and copyright

There have been an abundance of lawsuits over the use of protected works to train generative AI models, and the UK government has just initiated a new consultation on these topics through the Intellectual Property Office.

Use of protected works to train AI

The government recognises the problems of using protected works to train generative AI models. It anticipates that the Getty Images vs Stability AI case going through the High Court will resolve this. This case centres around the unauthorised use of images owned by Getty to automatically produce images. Nonetheless, such a resolution will take several years. In the meantime, the government has proffered four options. The first three options are as follows:

• Do nothing
• Strengthen copyright to require a licence before using the works
• Introduce a broad exception to allow data mining and training of generative AI.

The government is emphasising a fourth choice which is also being developed in the EU. This is to introduce an exception which allows data mining for any purpose, including commercial purposes. It would apply only where the user has lawful access to the relevant works. Additionally, it would apply where the rights owner, either individually or collectively, has not expressly reserved its rights. As part of that, the government is also evaluating if there is a need for greater standardisation of rights reservation protocols.

Developers would also have to be transparent about what content they are using to train their models. This would follow similar requirements under the EU AI Act and California’s Assembly Bill.

Ownership of AI-generated works

The government identifies “AI-assisted works” where a human creator makes use of an AI tool. These are already protected by copyright. It also recognises “computer-generated works” allowing for those created by AI without human intervention.

The consultation notes two potential issues with computer-generated works. First, “originality” is important. Like all copyrighted output, it means the “author’s own intellectual creation” with a human’s “personal touch”. This is absent with AI-generated content. Second, there is criticism that there should not be a separate category of protection for those which are computer-generated. Of course, AI-generated output which infringes existing rights would still be caught by the original copyright.

The government is considering three options although it does not favour the first two. One is to leave the law unchanged because it works well enough. Another is to clarify the law about originality. In fact, it favours the third choice. This would remove the protection of computer-generated works to be consistent with the USA and EU.

Removing this would leave such content in the public domain. On the positive side, this would prevent the build-up of non-original works. It would also avoid the issue of conflicting rights for similar output generated by similar tools.

What this means for developers, rights owners and users

This is a positive development and should provide clarity. Rights owners and AI developers should respond to this consultation to ensure their voices are heard. Until the outcome is published, AI developers need to beware of the risks of unauthorised use of protected works to train models. There are numerous lawsuits against generative AI developers. Even if an AI developer wins the lawsuit, defending it will take time, effort and money. This process will distract from the business.

Rights owners need to remain vigilant over unauthorised use of their creations. Again, winning a case is costly. It is better to try to keep tighter control of the works in the first place.

Users of such tools should review outputs. If these are covered by copyright, the rights owner might take action against them in addition to the developer.